【摘要】 为了使论文的被引次数跨学科可比需要用学科标准化方法消除论文被引次数指标上的学科差异。在评价论文集合层面的标准化表现时,线性标准化方法的线性变换特征使其比非线性方法更合适。在已有线性学科标准化方法中虽然比均值法应用最为广泛,但近期关于标准化方法标准化效果的研究显示,比均值法的效果显著但并不理想。意识到比均值法只是一般线性方法y=kx+b的特例,为了寻找效果更理想的线性标准化方法,本研究从线性方法的一般形式y=kx+b和学科标准化的理想目标——不同学科标准化的引文分布近似同分布出发建立了求解更有效的线性标准化方法的最优化模型并导出了基于最优化模型的线性标准化方法。标准化效果的比较显示,基于不加权优化模型的线性标准化方法的效果沿引文分布失衡,基于加权优化模型的缩放法y=kx相对于比均值法有微弱的优势而仿射法y=kx+b显著优于比均值法。在此基础上本研究以39所“985工程”大学为样本实证了比均值法和基于加权优化模型的线性标准化方法下大学论文质量排名的变异情况,考察了基于加权优化模型的线性标准化方法在标准化效果上的优势对实际科研评价结果的影响,相关结论为实际科研评价中标准化方法的选择提供了客观依据。更多还原

【Abstract】 To make the citations of publications comparable across fields,the field differences in citations have to be eliminated by field normalization methods.The linearity of linear normalization methods makes them more appropriate than nonlinear ones when evaluating the normalized citation performance at the aggregate level.Though mean-based method is most widely used among existing linear normalizations methods,recent studies on the normalization effect showed that its normalization effect is remarkable but less than ideal.Realising that mean-based method is just a special case of the general linear method y = kx + b,starting from the form of general linear method and the ideal objective of field normalization—universality of normalized citation distributions,we established an optimization model to look for more effective linear normalization methods.We derived the optimization-based linear normalization methods(OPTLNM) and made a comparison of the normalization effect between OPTLNM and mean-based method.The results showed that while the normalization performance of unweighted-optimization-based linear method is unbalanced along the citation distributions,the weighted- optimization-based scaling method y- kx shows trivial advantage but affine method y = kx + b shows significant advantage over mean-based method.In order to investigate the influence of advantage in normalization effect on the results of research evaluation,taking the 39 universities in " Project 985 " as samples,we made an empirical study of the variation of university publication quality rankings obtained under different normalization methods.Relevant findings can serve as objective foundations to the choice of proper normalization methods in practical research evaluation.更多还原